Three-dimensional temperature and salinity field inversion method and system based on ocean temperature and salinity profile

Abstract

The embodiment of the application discloses a three-dimensional temperature and salinity field inversion method and system based on marine temperature and salinity profile. The method comprises the following steps: obtaining an observation profile file corresponding to an observation point, and extracting key information, the key information comprising longitude, latitude, time and temperature and salinity profile, the temperature and salinity profile comprising temperature profile and salinity profile; loading a pre-trained intelligent reconstruction model; when the observation point is located in a preset modeling area range, searching for a plurality of loaded intelligent reconstruction models closest to the observation point in the modeling area range to obtain an intelligent reconstruction model of a triangular net vertex; calculating data of a marine three-dimensional temperature and salinity field according to the intelligent reconstruction model of the triangular net vertex to obtain three-dimensional body data; calculating a marine three-dimensional temperature and salinity field centered on the observation profile by using an interpolation method combined with the three-dimensional body data; and saving three-dimensional temperature and salinity field data of a target point. By implementing the method of the embodiment of the application, three-dimensional temperature and salinity field reconstruction based on one-dimensional profile data can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of temperature and salinity field reconstruction, and more particularly to a three-dimensional temperature and salinity field inversion method and system based on ocean temperature and salinity profiles. BACKGROUND

[0002] The temperature and salinity of seawater are key physical parameters in ocean science research. They not only determine the density of seawater, but also play a crucial role in driving ocean dynamic processes. Many ocean phenomena, such as ocean currents, mesoscale eddies, and internal waves, often occur at specific depth regions below the sea surface, and their occurrence and development are closely related to the three-dimensional temperature and salinity structure of the ocean. Therefore, accurately obtaining the three-dimensional distribution of the ocean temperature and salinity field is crucial for a deep understanding of ocean dynamics and environmental changes. However, due to the complexity of the ocean environment itself and the limitations of data collection methods, there are many challenges in fully analyzing the three-dimensional temperature and salinity structure of the ocean.

[0003] Currently, the main methods for obtaining ocean temperature and salinity profiles include field measurements and remote sensing technology. For example, buoy profile observations can provide temperature and salinity variation data in the vertical direction, but their coverage is limited and it is difficult to capture spatial variations over a large area. On the other hand, satellite remote sensing technology can provide large-area sea surface temperature and salinity data, but it cannot directly obtain the three-dimensional temperature and salinity distribution of the ocean interior, limiting the understanding of deep-sea temperature and salinity fields. Therefore, three-dimensional temperature and salinity field reconstruction based on surface observation data has become an important research area. Traditional three-dimensional temperature and salinity field reconstruction methods usually rely on regression models or empirical orthogonal functions (EOF), which attempt to infer the entire ocean temperature and salinity profile from limited data points. Although some success has been achieved, the accuracy and applicability of these methods are limited when dealing with complex or variable ocean environments. In addition, methods based on surface parameters still face challenges in inferring the entire three-dimensional temperature and salinity field, as the amount of information provided by these methods is relatively small, making it difficult to fully describe the three-dimensional structure of the entire ocean area.

[0004] With the advancement of technology and the increase in data volume, artificial intelligence-based methods are increasingly being applied to the reconstruction of three-dimensional temperature and salinity fields. In particular, algorithms such as support vector machines, random forests, and convolutional neural networks have attracted attention due to their strong modeling capabilities. Although some studies have used these methods in combination with surface parameters to achieve a transition from two-dimensional to three-dimensional temperature and salinity fields, research on three-dimensional temperature and salinity field reconstruction based on one-dimensional profile data is still lacking. For example, Chinese patent CN115712807A introduces a method for inverting the three-dimensional temperature and salinity field of the ocean using a U-Net neural network model, which can preserve information at different convolution layers, thereby improving prediction accuracy. Although this method has made progress using satellite remote sensing data, there is still a gap in three-dimensional temperature and salinity field reconstruction based on one-dimensional profile data.

[0005] Therefore, it is necessary to design a new method to realize three-dimensional temperature and salinity field reconstruction based on one-dimensional profile data. SUMMARY

[0006] The present application aims to overcome the defects of the prior art, and provide a three-dimensional temperature and salinity field inversion method and system based on ocean temperature and salinity profile.

[0007] To achieve the above-mentioned purpose, the present application adopts the following technical scheme: a three-dimensional temperature and salinity field inversion method based on ocean temperature and salinity profile, comprising:

[0008] Obtain the observation profile file corresponding to the observation point, and extract the key information, wherein the key information includes latitude, longitude, time and temperature and salinity profile, and the temperature and salinity profile includes temperature profile and salinity profile;

[0009] Load the pre-trained intelligent reconstruction model;

[0010] When the observation point is located in the preset modeling area range, find the nearest several loaded intelligent reconstruction models to the observation point in the modeling area range to obtain the intelligent reconstruction model of the triangular net vertex;

[0011] According to the intelligent reconstruction model of the triangular net vertex, the data of the ocean three-dimensional temperature and salinity field is calculated to obtain the three-dimensional body data;

[0012] The ocean three-dimensional temperature and salinity field centered on the observation profile is calculated by using the interpolation method combined with the three-dimensional body data to obtain the three-dimensional temperature and salinity field data of the target point;

[0013] Save the three-dimensional temperature and salinity field data of the target point.

[0014] Further technical scheme thereof is that the training process of the pre-trained intelligent reconstruction model comprises:

[0015] Construct the triangular net of the modeling area and construct the intelligent reconstruction model at the vertex;

[0016] Train the intelligent reconstruction model using reanalysis data set.

[0017] Further technical scheme thereof is that the training of the intelligent reconstruction model using reanalysis data set comprises:

[0018] Select the HYCOM reanalysis data set to obtain the training set;

[0019] According to the vertex geographic coordinates of the triangular net of the modeling area, the corresponding three-dimensional temperature and salinity field data is extracted from the training set and matched with the vertex position to obtain the extraction result;

[0020] constructing a data set for model training and verification for each vertex using the extraction result;

[0021] training the intelligent reconstruction model by combining the training and verification data sets using an MSE loss function and optimizing parameters through a back propagation algorithm.

[0022] A further technical solution is that the intelligent reconstruction model calculates the data of the ocean three-dimensional temperature and salinity field according to the triangular mesh vertex, to obtain three-dimensional volume data, including:

[0023] The temperature and salinity profile and the time are input into the intelligent reconstruction model of the triangular mesh vertex to calculate the data of the ocean three-dimensional temperature and salinity field, to obtain three-dimensional volume data.

[0024] A further technical solution is that the intelligent reconstruction model includes a deep neural network model with a double-branch structure, the deep neural network model with a double-branch structure includes a temperature and salinity profile processing branch, a time processing branch, a feature integration layer, and a reconstruction module; the temperature and salinity profile processing branch includes a one-dimensional convolution layer; the time processing branch includes a fully connected layer; and the reconstruction module includes a fully connected layer.

[0025] A further technical solution is that the time processing branch uses a ReLU activation function, a first fully connected layer expands three input features to eight output features, and a second fully connected layer further expands to 48 output features, matching the output dimension of the temperature and salinity profile processing branch.

[0026] A further technical solution is that the temperature and salinity profile and the time are input into the intelligent reconstruction model of the triangular mesh vertex to calculate the data of the ocean three-dimensional temperature and salinity field, to obtain three-dimensional volume data, including:

[0027] extracting spatial features from the temperature and salinity profile in the vertical direction using a one-dimensional convolution layer to obtain profile features;

[0028] processing the input time information through a fully connected layer to obtain time features;

[0029] integrating the time features and the profile features through addition to obtain an integration result;

[0030] processing the integration result through a fully connected layer and reshaping it into a 3D field form, applying a maximum and minimum value normalization method for conversion to obtain actual temperature and salinity values, forming three-dimensional volume data.

[0031] A further technical solution is that the three-dimensional volume data is calculated using an interpolation method to obtain the three-dimensional temperature and salinity field data of the target point, including:

[0032] determining a weight coefficient of the observation point in the triangulation network using a barycentric interpolation method;

[0033] based on the three-dimensional body data and the weight coefficient, a three-dimensional temperature and salinity field at the observation point is comprehensively estimated by using an interpolation method to obtain three-dimensional temperature and salinity field data of the target point.

[0034] Further technical solutions are as follows: based on the three-dimensional body data and the weight coefficient, a three-dimensional temperature and salinity field at the observation point is comprehensively estimated by using an interpolation method to obtain three-dimensional temperature and salinity field data of the target point, including:

[0035] weighting and summing the weight coefficient of each vertex and the corresponding three-dimensional body data to obtain three-dimensional temperature and salinity field data of the target point.

[0036] The application also provides a three-dimensional temperature and salinity field inversion system based on a marine temperature and salinity profile, including:

[0037] An acquisition unit is configured to acquire an observation profile file corresponding to an observation point and extract key information, wherein the key information includes longitude and latitude, time, and a temperature and salinity profile, and the temperature and salinity profile includes a temperature profile and a salinity profile.

[0038] A loading unit is configured to load a pre-trained intelligent reconstruction model.

[0039] A searching unit is configured to, when the observation point is located within a preset modeling area range, search for a plurality of loaded intelligent reconstruction models closest to the observation point in the modeling area range to obtain an intelligent reconstruction model of a triangulation network vertex.

[0040] A calculation unit is configured to calculate data of a marine three-dimensional temperature and salinity field according to the intelligent reconstruction model of the triangulation network vertex to obtain three-dimensional body data.

[0041] An interpolation unit is configured to calculate a marine three-dimensional temperature and salinity field centered on the observation profile by using an interpolation method in combination with the three-dimensional body data to obtain three-dimensional temperature and salinity field data of the target point.

[0042] A saving unit is configured to save the three-dimensional temperature and salinity field data of the target point.

[0043] Compared with the prior art, the application has the following beneficial effects: by acquiring one-dimensional temperature and salinity profile data of an observation point, the application finds the closest triangulation network vertex model in the modeling area by using a pre-trained intelligent reconstruction model, calculates three-dimensional temperature and salinity field data at these vertices, and reconstructs an accurate three-dimensional temperature and salinity field centered on the observation point by using an interpolation method in combination with these data, and finally saves the reconstruction result; in this way, efficient conversion and reconstruction from limited one-dimensional profile data to detailed three-dimensional temperature and salinity field are realized.

[0044] The application will be further described below with reference to the drawings and specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0046] Figure 1 The application scenario schematic diagram of the three-dimensional temperature and salinity field inversion method based on the ocean temperature and salinity profile provided by the embodiments of the present application is shown in the figure.

[0047] Figure 2 The flowchart of the three-dimensional temperature and salinity field inversion method based on the ocean temperature and salinity profile provided by the embodiments of the present application is shown in the figure.

[0048] Figure 3 The sub-flowchart of the three-dimensional temperature and salinity field inversion method based on the ocean temperature and salinity profile provided by the embodiments of the present application is shown in the figure.

[0049] Figure 4 The sub-flowchart of the three-dimensional temperature and salinity field inversion method based on the ocean temperature and salinity profile provided by the embodiments of the present application is shown in the figure.

[0050] Figure 5 The schematic diagram of the triangular net of the modeling area provided by the embodiments of the present application is shown in the figure.

[0051] Figure 6 The schematic block diagram of the three-dimensional temperature and salinity field inversion system based on the ocean temperature and salinity profile provided by the embodiments of the present application is shown in the figure.

[0052] Figure 7 The schematic block diagram of the computer device provided by the embodiments of the present application is shown in the figure. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor belong to the scope of protection of the present application.

[0054] It should be understood that when used in the specification and the appended claims, the terms "comprise" and "include" indicate the presence of described features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0055] It should also be understood that the terms used herein are for the purpose of describing particular embodiments and are not intended to limit the application. As used in the specification and the appended claims, the singular forms "a," "an" and "the" are intended to include plural forms as well, unless the context clearly indicates otherwise.

[0056] It should be further understood that the term "and / or" used in the specification and the appended claims means one or more of the associated listed items as well as all possible combinations of the items and includes the combinations.

[0057] Reference is made to Figure 1 and Figure 2 , Figure 1 The application scenario schematic diagram of the three-dimensional temperature and salinity field inversion method based on the ocean temperature and salinity profile provided by the embodiments of the application. Figure 2 The schematic flowchart of the three-dimensional temperature and salinity field inversion method based on the ocean temperature and salinity profile provided by the embodiments of the application. The three-dimensional temperature and salinity field inversion method based on the ocean temperature and salinity profile is applied to a server, which performs data interaction with a terminal, and uses a pre-trained intelligent reconstruction model to perform a three-dimensional temperature and salinity field inversion method. Specifically, first, the temperature and salinity profile (including temperature and salinity information), latitude and longitude, and time of the observation point are obtained, and the intelligent reconstruction model suitable for the region is loaded; then, the data of the intelligent reconstruction model on the triangle net vertex closest to the observation point is calculated to obtain the preliminary three-dimensional body data; then, the temperature and salinity profile and the time are input into the model, and after deep neural network processing (including one-dimensional convolution layer to extract spatial features, full connection layer to process time information, and integration of the two features), detailed three-dimensional temperature and salinity field data centered on the observation point are generated; finally, the three-dimensional temperature and salinity field data at the target point are comprehensively estimated by using an interpolation method combined with a weight coefficient. The entire process effectively realizes accurate reconstruction from one-dimensional profile data to three-dimensional temperature and salinity field by means of a deep neural network model with a double-branch structure.

[0058] Figure 2 The flowchart of the three-dimensional temperature and salinity field inversion method based on the ocean temperature and salinity profile provided by the embodiments of the application. As shown in Figure 2 the method comprises the following steps S110 to S160.

[0059] S110, an observation profile file corresponding to an observation point is obtained, and key information is extracted, wherein the key information includes latitude and longitude, time, and a temperature and salinity profile, and the temperature and salinity profile includes a temperature profile and a salinity profile.

[0060] In this embodiment, step S110 is the starting point of the entire method of inverting three-dimensional temperature and salinity fields based on ocean temperature and salinity profiles. The main purpose is to obtain the observation profile file corresponding to the observation point and extract key information from it. This process is crucial for the success of subsequent steps, as it provides the basic data needed to build and train intelligent reconstruction models.

[0061] Specifically, step S110 includes the following aspects:

[0062] Obtain the observation profile file: First, we need to obtain the observation profile file of a specific observation point from the ocean monitoring equipment or database. These files usually include data measured by CTD (conductivity, temperature, depth) probes and other instruments at different depths in a certain location, covering temperature and salinity values at a series of depth levels.

[0063] Extract latitude and longitude information: Extract the specific geographic location of the observation point, i.e. longitude and latitude, from the observation profile file. This step is to determine the exact location of the observation point in the study area, so as to associate it with the vertices of the triangular network constructed in subsequent steps.

[0064] Record the timestamp: In addition to spatial coordinates, we also need to accurately record the time information of the observation, including date and time (such as year, month, day, hour). Time information is very important for understanding the dynamic characteristics of ocean environmental changes, as it helps us consider factors such as seasonal changes, diurnal cycles, and their impact on the distribution of ocean temperature and salinity.

[0065] Parse temperature and salinity profile data: Finally, and most importantly, extract temperature profile and salinity profile data from the observation profile file. These data describe the temperature and salinity changes at different depth levels from the sea surface to the seabed, and are the core basis for training intelligent reconstruction models and ultimately reconstructing three-dimensional temperature and salinity fields.

[0066] In summary, step S110 systematically collects and processes the above key information, laying a solid foundation for subsequent steps. This not only helps ensure the accuracy of model training, but also makes it possible to achieve high-precision three-dimensional temperature and salinity field reconstruction. This step emphasizes the importance of data quality and accuracy for the entire technical solution.

[0067] S120, load the pre-trained intelligent reconstruction model.

[0068] In this embodiment, the intelligent reconstruction model refers to a pre-trained deep learning model designed to model the relationship between known temperature-salinity profile data (i.e., temperature and salinity as a function of depth) and three-dimensional temperature-salinity fields. Specifically, this model is trained using high-resolution HYCOM reanalysis datasets to accurately reconstruct ocean temperature-salinity fields within a specific sea area.

[0069] The model structure is as follows:

[0070] Dual-branch structure: This model typically employs a deep neural network with a dual-branch structure, where one branch processes temperature-salinity profile data in the vertical direction, and the other branch handles temporal information.

[0071] Input layer: Accepts temperature-salinity profile data from the vertices of the triangular mesh in the selected modeling area as input.

[0072] Hidden layers: Composed of multiple layers of neurons, each containing a certain number of nodes connected by nonlinear activation functions (such as ReLU) to extract features from the input data.

[0073] Output layer: Produces temperature and salinity predictions for each location in three-dimensional space.

[0074] A triangular mesh of the study area is constructed using the Delaunay triangulation method, and corresponding three-dimensional temperature-salinity field data are extracted from the HYCOM reanalysis dataset. The prepared data are used as input, and the model parameters are learned and optimized by minimizing the mean squared error (MSE) loss function using the backpropagation algorithm. The model performance is evaluated on an independent validation set, and the model architecture or hyperparameters are adjusted based on the results until satisfactory prediction accuracy is achieved.

[0075] This intelligent reconstruction model can be applied in various fields such as marine scientific research, climate change monitoring, and fishery resource management. It can help researchers better understand and predict the trends of changes in the marine environment, especially in areas where observational data is scarce or difficult to obtain.

[0076] In summary, the intelligent reconstruction model is a highly specialized machine learning tool designed to improve the understanding and prediction of complex marine physical phenomena. By combining advanced mathematical modeling techniques with large amounts of observational data, such models provide strong support for exploring the mysteries of the ocean.

[0077] The training process of the pre-trained intelligent reconstruction model includes:

[0078] Constructing a triangular mesh of the modeling area and constructing an intelligent reconstruction model at the vertices;

[0079] The intelligent reconstruction model is trained using a reanalysis dataset.

[0080] Specifically, the training of the intelligent reconstruction model using the reanalysis dataset comprises:

[0081] A HYCOM reanalysis dataset is selected to obtain a training set.

[0082] Corresponding three-dimensional temperature and salinity field data are extracted from the training set according to the vertex geographic coordinates of the triangular mesh of the modeling area and matched with the vertex positions to obtain an extraction result.

[0083] The extraction result is used to construct a data set for model training and verification for each vertex.

[0084] The intelligent reconstruction model is trained using the training and verification data sets with an MSE loss function and optimized parameters through a backpropagation algorithm.

[0085] Specifically, the specific sea area range to be studied is first determined. This step may need to be adjusted according to actual needs, such as specific oceanographic problems or geographical features,

[0086] A triangular mesh covering the entire area is constructed using the Delaunay triangulation method, as shown in Figure 5 This algorithm maximizes the minimum angle principle to ensure that the generated triangles are as close to equilateral triangles as possible, thereby providing more accurate spatial interpolation results. This method ensures that the generated triangular mesh has the maximum angle optimization property, thereby improving the accuracy and stability of spatial interpolation analysis.

[0087] At each triangular mesh vertex, an intelligent reconstruction model is established based on the mapping relationship between the known temperature and salinity profile and the three-dimensional temperature and salinity field. That is, at each triangular mesh vertex, a deep learning model is established based on the relationship between the known temperature and salinity profile data (i.e., temperature and salinity data varying with depth) and the three-dimensional temperature and salinity field. These models typically include the aforementioned dual-branch structure deep neural network, which processes vertical profile data and temporal information separately.

[0088] These models will serve as core components in subsequent steps for inferring three-dimensional temperature and salinity fields at any location.

[0089] To ensure the amount and quality of model training data, the HYCOM (Hybrid Coordinate Ocean Model) reanalysis dataset is selected as the main data source. HYCOM provides high-resolution global ocean temperature and salinity field data, which is suitable for training deep learning models.

[0090] Based on the geographic coordinates of the vertices of the triangular mesh within the modeling area, corresponding 3D temperature and salinity field data are extracted from the HYCOM dataset and precisely matched with the vertex locations. This step ensures the authenticity and accuracy of the training data.

[0091] Using the extraction results described above, specialized training and validation sets are constructed for each vertex of the triangulation network. These datasets not only contain spatial distribution information of temperature and salinity but also take into account the influence of the time dimension.

[0092] Mean squared error (MSE) is used as the loss function to measure the difference between the model output and the actual data, and a backpropagation algorithm is applied to optimize the model parameters. The backpropagation algorithm updates the model parameters to reduce the gap between the predicted and true values. This process involves calculating gradients and adjusting weights according to gradient descent until a preset stopping condition (such as the number of iterations, convergence threshold, etc.) is reached. This process is repeated iteratively until the model achieves satisfactory prediction accuracy.

[0093] The process of loading a pre-trained intelligent reconstruction model involves the following aspects:

[0094] The first step is to read the pre-trained intelligent reconstruction model files from the storage medium. These files typically contain information such as the model architecture definition and weight parameters.

[0095] The read model file is loaded into memory, and necessary initialization operations are performed, such as setting the computing device (CPU or GPU) for model runtime and adjusting the input / output interfaces.

[0096] Before it is used in practice, the loaded model needs to be tested or verified to ensure that its structure is intact and its functions are normal.

[0097] Through the steps described above, pre-trained intelligent reconstruction models can be rapidly deployed and applied to new observation points or study areas, significantly improving the efficiency and accuracy of three-dimensional temperature and salinity field inversion. This technical solution not only overcomes the limitations of traditional methods but also provides strong data support for marine scientific research.

[0098] S130. When the observation point is located within a preset modeling area, find several loaded intelligent reconstruction models that are closest to the observation point within the modeling area to obtain the intelligent reconstruction model of the triangular mesh vertices.

[0099] In this embodiment, the smart reconstruction model at a triangulation point refers to a pre-trained deep learning model that is deployed at each vertex of a regional triangulation network constructed using the Delaunay triangulation method. Each smart reconstruction model aims to predict the three-dimensional temperature-salinity field at that vertex location based on the ocean temperature-salinity profile data (i.e., temperature and salinity as a function of depth).

[0100] First, the exact geographical location of the observation point is determined. This usually involves receiving actual measurement data from sensors or other data collection devices. Once the location of the observation point is determined, the system will search for the nearest few triangulation vertices within the pre-set modeling area. Here, "nearest" generally refers to the shortest geographical distance, but it can also be defined as other forms of distance or similarity measures according to specific application requirements. For each vertex found, the system will load the smart reconstruction model that has been pre-trained and deployed on it. These models are obtained by training on historical temperature-salinity profile data and corresponding three-dimensional temperature-salinity field data, and can accurately infer the corresponding three-dimensional temperature-salinity distribution from the input temperature-salinity profile.

[0101] The smart reconstruction model includes a deep neural network model with a dual-branch structure, which includes a temperature-salinity profile processing branch, a time processing branch, a feature integration layer, and a reconstruction module. The temperature-salinity profile processing branch includes a one-dimensional convolution layer. The time processing branch includes a fully connected layer. The reconstruction module includes a fully connected layer.

[0102] Specifically, the smart reconstruction model is a deep learning model specifically designed for reconstructing three-dimensional temperature-salinity fields from ocean temperature-salinity profile data. This model adopts a unique dual-branch structure, which can effectively process spatial (temperature-salinity profile) and temporal dimension information, thus providing accurate three-dimensional temperature-salinity field prediction. Specifically, it includes:

[0103] Temperature-salinity profile processing branch:

[0104] Input layer: receives temperature-salinity profile data in the vertical direction as input. These data usually include a series of measured values of temperature and salinity as a function of depth.

[0105] One-dimensional convolution layer (Conv1D): composed of eight cascaded Conv1D layers, each layer has a kernel size of 3, padding of 1, step of 1, and uses ReLU activation function. This setting helps the model capture spatial features on the temperature-salinity profile, such as local trends or patterns.

[0106] Concatenation operation: To ensure that no original information is lost, the input profile is concatenated with the output of the last layer of Conv1D, forming the final profile branch output. This process allows the model to utilize not only the high-level features extracted by the convolutional layers but also to retain the detailed information from the original input.

[0107] Temporal processing branch:

[0108] Input layer: Receives temporal information representing the observation time (e.g., month, day, and hour) as input.

[0109] Fully connected layers (FC): Composed of two cascaded fully connected layers, the first layer has 3 input features (month, day, hour) and 8 output features; the second layer has 8 input features and 48 output features. Each layer is followed by a ReLU activation function. This part of the design aims to convert the temporal information into a feature representation that is useful for the model, helping to understand the impact of seasonality and diurnal variations on ocean temperature and salinity fields.

[0110] Specifically, the temporal processing branch employs ReLU activation functions, with the first fully connected layer expanding three input features to eight output features, and the second fully connected layer further expanding to 48 output features, matching the output dimension of the temperature-salinity profile processing branch.

[0111] Feature integration layer: In the feature integration layer, the output features from the temperature-salinity profile processing branch and the temporal processing branch are combined through a simple addition operation. This not only simplifies the model structure but also allows for easy removal of a branch for ablation studies without affecting the functionality of other parts.

[0112] Reconstruction module:

[0113] Fully connected layers (FC): Contains a fully connected layer that accepts 48 features from the feature integration layer as input and outputs 2904 (24x11x11) features. The goal of this step is to recover the complete three-dimensional temperature and salinity fields from the integrated features.

[0114] Finally, the output of the fully connected layer is reshaped into a three-dimensional field format, covering the temperature and salinity distribution of the entire water body. Subsequently, the maximum and minimum value normalization method is applied to adjust the numerical range, making it conform to the actual physical meaning.

[0115] The temporal processing branch plays a crucial role in the intelligent reconstruction model, primarily responsible for processing and extracting time-related features. This branch employs a fully connected layer (FC) structure and uses ReLU activation functions to increase non-linear capabilities, enabling it to better capture complex patterns in the input data. The specific architecture is detailed as follows:

[0116] The input features of the time processing branch are time information, which usually includes data of three dimensions: month, day, and hour, representing the time point of observation. These three dimensions serve as the input features of the first fully connected layer.

[0117] The first fully connected layer (FC1) information is as follows:

[0118] Input size: 3 (three time features: month, day, and hour)

[0119] Output size: 8 (the input features are expanded to 8 output features through this layer)

[0120] Activation function: ReLU (Rectified Linear Unit), which is used to introduce nonlinearity, allowing the model to learn more complex representations. In this layer, the input time features are transformed by a weight matrix, then added to a bias term, and then nonlinearly converted by the ReLU function to generate an 8-dimensional feature vector with richer representation ability.

[0121] The second fully connected layer (FC2) information is as follows:

[0122] Input size: 8 (output feature number from the first fully connected layer)

[0123] Output size: 48 (further expand the feature dimension to 48 to match the output dimension of the temperature-salinity profile processing branch)

[0124] Activation function: ReLU activation function is also used to continue to enhance the model's representation ability and learning ability of complex patterns. The role of this layer is to further process the 8-dimensional feature vector generated by the previous layer and map it to a higher-dimensional space (i.e., 48 dimensions). This not only helps to improve the model's representation ability, but also lays the foundation for subsequent integration of features with the temperature-salinity profile processing branch.

[0125] The 48-dimensional feature vector output by the time processing branch will be merged with the output of the temperature-salinity profile processing branch. This design ensures that information from two different data sources can be effectively integrated in the same dimension, making full use of spatial and temporal information to reconstruct the three-dimensional temperature-salinity field.

[0126] The time processing branch, through the design of two fully connected layers, first converts the original time information (month, day, and hour) into higher-level abstract features, and then further expands the dimensions of these features to match the output of the temperature-salinity profile processing branch. Such architecture not only ensures the effective use of time information, but also provides convenience for subsequent feature integration, and is an important part of achieving accurate ocean environment simulation.

[0127] In summary, the intelligent reconstruction model achieves efficient and accurate reconstruction of the three-dimensional temperature and salinity field by combining the carefully designed double-branch structure, the profile processing branch responsible for analyzing spatial information, the time processing branch responsible for analyzing temporal dependence, and the effective combination of feature integration layer and reconstruction module. This method not only improves the spatial resolution of traditional methods, but also enhances the adaptability of the model to different environmental conditions.

[0128] S140, calculating the data of the three-dimensional temperature and salinity field according to the intelligent reconstruction model of the triangular mesh vertices to obtain three-dimensional volume data.

[0129] In this embodiment, the three-dimensional volume data refers to a data set covering the temperature and salinity distribution of the entire water space (usually a specific sea area) calculated by the model. This data set is represented in the form of a three-dimensional matrix, where each element represents the temperature or salinity value at a specific location (x, y, z coordinates). In this invention, after the full connection layer processing, the output is reshaped into a 24x11x11 three-dimensional matrix, and converted into actual physical quantities (such as Celsius or PSU) through maximum and minimum value normalization to form the final three-dimensional temperature and salinity field.

[0130] In an embodiment, referring to Figure 3 The above step S140 can include steps S141-S144.

[0131] S141, extracting spatial features from the temperature and salinity profile in the vertical direction using a one-dimensional convolution layer to obtain profile features.

[0132] In this embodiment, the profile feature refers to the spatial feature extracted from the temperature and salinity profile data in the vertical direction. These features are generated by a series of one-dimensional convolution layer (Conv1D) operations, which are used to capture the temperature or salinity variation patterns at different depths on the ocean profile. Specifically, the profile branch consists of eight cascaded Conv1D layers, each using a kernel size of 3, padding of 1, step of 1, and ReLU activation function. These layers can effectively extract local spatial information and mine global spatial information through multi-layer stacking. In addition, the original input and the output of the last Conv1D layer are spliced together as the final profile feature, ensuring that all important information is retained.

[0133] In this stage, the input temperature or salinity profile data is processed through a series of one-dimensional convolutional layers (Conv1D). Each convolutional layer uses a kernel size of 3, padding of 1, and a stride of 1, with a ReLU activation function. These parameter settings ensure that local spatial information in the profile data can be effectively captured. After multiple convolution operations, the original input and the convolutional layer output are connected as the final profile feature output. This design not only enhances the model's understanding of global spatial information but also ensures that the original information is not lost, thereby improving the reconstruction accuracy.

[0134] S142, processing the input time information through a fully connected layer to obtain time features.

[0135] In this embodiment, time features refer to the feature vector obtained by processing the input time information (such as month, date, hour, etc.). The time branch contains two fully connected layers, the first layer expands the input dimension from 3 to 8, and the second layer further expands to 48, followed by a ReLU activation function after each layer. This method allows the model to learn time-related patterns, such as how seasonal and diurnal variations affect ocean temperature and salinity. Therefore, time features provide an overall background information similar to climate averages, which helps improve prediction accuracy.

[0136] For the time branch, first receive the time information such as month, date, and hour of the year as input. This branch consists of two cascaded fully connected layers, where the first fully connected layer expands the input dimension from 3 to 8, and the second fully connected layer further expands the dimension to 48. Each layer is followed by a ReLU activation function to introduce non-linear factors. This process allows the model to learn time-related patterns, such as how seasonal and diurnal variations affect ocean temperature and salinity distribution. The time features generated in this way provide an overall background information similar to climate averages, which helps improve prediction accuracy.

[0137] S143, integrating the time features and the profile features through addition to obtain an integrated result.

[0138] In this embodiment, the integrated result is the output after combining the profile features and the time features. To effectively combine these two types of information, the invention uses an addition operation instead of a direct connection method. This means that the profile features and the time features are first converted into vectors of the same dimension, and then they are combined through a simple addition operation. This not only simplifies the model structure but also maintains the consistency of subsequent modules, making it easier to analyze the role of each component. The integrated result is then passed to the next fully connected layer to generate the final three-dimensional temperature and salinity field.

[0139] To effectively combine spatial and temporal information, the present application employs a simple addition operation to integrate profile features and temporal features. Compared to direct concatenation, this method maintains the structure of subsequent modules unchanged, simplifying the model complexity. At the same time, it is also easier to analyze the role of each component when conducting ablation studies. This additive integration method allows the model to flexibly utilize information from different sources, thereby more accurately predicting marine environmental variables.

[0140] S144, the integration result is processed through a fully connected layer and reshaped into a 3D field form, and a max-min normalization method is applied for conversion to obtain actual temperature and salinity values, forming a three-dimensional volume data.

[0141] The last step is to pass the integrated features obtained in the previous steps to a fully connected layer with an output dimension of 2904 (i.e., 24x11x11), corresponding to the spatial resolution of the target three-dimensional temperature and salinity field. Then, this high-dimensional vector is reshaped into a three-dimensional matrix form, and the max-min normalization method is used to convert it back to the physical meaning of temperature and salinity values. This step ensures that the model output can be directly used for scientific research or other application scenarios, providing intuitive and easy-to-understand result display.

[0142] In summary, through the above steps, the present embodiment successfully achieves the goal of inverting the three-dimensional temperature and salinity field of the ocean based on single-point observation profile data, significantly improving the precision and efficiency of the existing technology. At the same time, this method also has good universality and adaptability, suitable for temperature and salinity field reconstruction tasks in different sea areas and conditions.

[0143] S150, using an interpolation method to combine the three-dimensional volume data to calculate the three-dimensional temperature and salinity field of the ocean centered on the observation profile to obtain the target point three-dimensional temperature and salinity field data.

[0144] In this embodiment, the target point three-dimensional temperature and salinity field data refers to the spatial distribution data of temperature (Thermohaline, abbreviated as Temperature) and salinity (Salinity) at a specific observation point position obtained through the above steps S151 and S152. Specifically, these data describe the thermodynamic properties and salinity concentration of the observation point in its surrounding environment, including the variation at different depth levels.

[0145] This three-dimensional temperature and salinity field data is of great significance for understanding the internal structure, dynamic processes, and interactions between the ocean and the atmosphere. It not only contains the temperature and salinity information at a certain observation point at a certain time, but also reflects their trends with depth, and is the basic data for studying ocean circulation, mixed layer depth, thermohaline circulation, etc.

[0146] Therefore, accurately estimating the target point three-dimensional temperature and salinity field data, i.e. calculating the precise temperature and salinity information at the observation point by combining the known three-dimensional body data with the interpolation method, is of great significance for improving the prediction accuracy of the ocean model, optimizing resource management strategies, and responding to climate change. This process ensures that even in areas with sparse data or difficult to measure directly, relatively reliable temperature and salinity field estimates can be obtained.

[0147] By using the interpolation method, the three-dimensional temperature and salinity field data at the known triangular net vertices are combined with the positional relationship of the observation point to estimate the three-dimensional temperature and salinity field data of the target point (i.e. the observation point) centered on the observation profile. This process not only considers the influence of spatial position, but also integrates time information, making the final temperature and salinity field data have high precision and reliability.

[0148] In an embodiment, referring to Figure 4 , the above step S150 can include steps S151-S152.

[0149] S151, using the barycentric interpolation method to determine the weight coefficient of the observation point in the triangular net.

[0150] In this embodiment, the barycentric interpolation method is a commonly used geometric interpolation technique, especially suitable for estimating the attribute value of a point inside a triangle. This method assigns weight coefficients based on the proportion of geometric area, and the specific steps are as follows:

[0151] First, determine which triangle composed of three vertices the observation point is located in according to its geographic coordinates.

[0152] Assuming that the three vertices of the triangle are A, B, and C, and the observation point to be interpolated is P, then the weight of P relative to A, B, and C is , , ; calculated by the following formula: ; ; . Where Area represents the calculated area, points A, B, and C are the three vertices of the triangle, and point P is a point inside the triangle, here is the point to be interpolated, , , are the interpolation coefficients of the three vertices.

[0153] These weight coefficients represent the importance or contribution of the observation point P relative to the three vertices of the triangle ABC.

[0154] S152, based on the three-dimensional body data and the weight coefficient, the three-dimensional temperature and salinity field at the observation point is comprehensively estimated by using the interpolation method to obtain the three-dimensional temperature and salinity field data of the target point.

[0155] In this embodiment, the weight coefficients of each vertex are weighted and summed with the corresponding three-dimensional body data to obtain the target point three-dimensional temperature and salinity field data.

[0156] After obtaining the weight coefficients of the observation point relative to the triangle vertices, the next step is to use these weight coefficients and the three-dimensional body data at the vertices (i.e., the three-dimensional distribution of temperature and salinity) to estimate the three-dimensional temperature and salinity field data at the observation point by weighted summation method.

[0157] Obtain the three-dimensional temperature and salinity field data at each vertex from the intelligent reconstruction model. This includes the distribution of temperature and salinity at different depth levels.

[0158] For temperature and salinity, the following operations are performed respectively: ; wherein, 、 、 are the interpolation coefficients of the three vertices, 、 、 are the three-dimensional temperature or salinity fields of points A, B, and C, represents the three-dimensional temperature or salinity field of point P.

[0159] Through the above steps, the three-dimensional temperature and salinity field data at the observation point can be accurately estimated, ensuring the spatial continuity and physical reasonableness of the results. This method effectively combines local details and global information, improving the accuracy and efficiency of three-dimensional temperature and salinity field reconstruction, and is particularly suitable for application in ocean dynamics and environmental change research.

[0160] S160, save the target point three-dimensional temperature and salinity field data.

[0161] In this embodiment, the target point three-dimensional temperature and salinity field data is saved as an nc file.

[0162] For the method of the present embodiment, first, select some points in the selected study area as the base vertices of the Delaunay triangulation. These vertices can be encrypted as needed to optimize the grid structure. Delaunay triangulation is a triangulation algorithm that maximizes the minimum angle, ensuring that the generated triangular mesh is stable and suitable for spatial analysis. At each vertex, a deep learning model is established based on the mapping relationship between the temperature-salinity profile and the three-dimensional temperature-salinity field. This model accepts temperature-salinity profile data at the vertex as input and outputs the three-dimensional temperature-salinity field at that location; HYCOM reanalysis dataset is used as the training data source. This dataset provides high-resolution global ocean temperature-salinity field data, covering a wide range of time spans and geographic ranges, making it ideal for training deep learning models. According to the geographic coordinates of the known vertices, the corresponding HYCOM data is extracted and matched with the vertex location to construct the training set and validation set. The mean squared error (MSE) loss function is used to evaluate the difference between the model's predicted values and the actual data, and the model parameters are optimized through the backpropagation algorithm. For each triangular mesh vertex, an intelligent reconstruction model for temperature and salinity is trained separately. To effectively extract spatial information from temperature-salinity profile data, the model uses multiple cascading one-dimensional convolution layers (Conv1D). These layers can capture important features in the vertical profile and enhance non-linear expression capabilities through rectified linear unit (ReLU) activation layers. Considering the importance of time factors on temperature-salinity field reconstruction, the month, date, and hour of the year are used as additional inputs to provide seasonal and diurnal variation-related background information.

[0163] Specifically, the temperature-salinity profile processing branch is responsible for processing the vertical profile data of temperature or salinity. It includes eight cascading Conv1D layers, each with a kernel size of 3, a padding of 1, and a stride of 1. The profile branch receives the original profile data as input and connects it with the output of the Conv1D layers to ensure that all key information is preserved. Finally, the profile branch produces an array containing 48 elements, effectively expanding the representation range of the profile information.

[0164] The time processing branch focuses on processing time information, including month, date, and hour. This branch consists of two fully connected layers, with the first layer converting 3 time features into 8 features and the second layer further expanding to 48 features. ReLU activation layers are applied between each layer to introduce nonlinearity. The output of the time branch is then added to the output of the profile branch, integrating spatiotemporal information together.

[0165] The reconstruction module receives the integrated features (a total of 48) from the extraction module and maps them to 2904 features through a single fully connected layer (corresponding to a spatial resolution of 24x11x11). This step aims to recover the complete three-dimensional temperature and salinity field distribution from the integrated features. Finally, the generated 3D field is adjusted to the actual temperature and salinity value range using the max-min normalization method.

[0166] This innovative dual-branch structure design not only allows the model to accurately capture complex patterns in the vertical profile, but also flexibly incorporates the effects of the time dimension, significantly improving the accuracy and efficiency of three-dimensional temperature and salinity field reconstruction. In addition, by using addition instead of concatenation to fuse the outputs of the profile and time branches in the extraction module, the model structure is more concise and easier to conduct ablation studies.

[0167] According to the geographic coordinates of the collected temperature and salinity profile, determine which triangular mesh unit it is located in, and identify the corresponding intelligent reconstruction model at the three vertices of the unit. Each vertex is equipped with 3 trained temperature reconstruction models and 3 salinity reconstruction models.

[0168] After standardizing the temperature and salinity profile of the observation point, it is input into the determined 6 vertex intelligent reconstruction models. Each vertex model outputs the corresponding three-dimensional temperature field or salinity field. In this way, we obtain the three-dimensional temperature field of the three vertices and the three-dimensional salinity field of the three vertices.

[0169] Using the barycentric interpolation method in geometric calculation, calculate the weight coefficient of the observation point relative to the triangular mesh vertices. Combining the obtained vertex three-dimensional temperature field and salinity field and the interpolation coefficient calculated in step S5, we can comprehensively obtain the three-dimensional temperature and salinity field of the observation point.

[0170] The method of this embodiment significantly surpasses the traditional reconstruction method based on surface data, and realizes efficient conversion from limited one-dimensional information to three-dimensional structure by using vertical profile data. Through systematic evaluation, this method shows reliability in practical application and can accurately reproduce the three-dimensional temperature and salinity distribution of the ocean, providing more accurate data support for ocean dynamics research and environmental monitoring.

[0171] In addition, the triangular mesh interpolation method proposed in this embodiment, combined with the regional triangular mesh reconstruction model constructed by numerical model data, can quickly obtain three-dimensional temperature and salinity field data at any location. This innovation not only improves the reconstruction accuracy, but also solves the problem of large computational resource consumption and long modeling time in traditional methods. Overall, the algorithm provided by the present invention has the advantages of simple operation, high efficiency, and strong adaptability, and is more flexible and efficient than traditional two-dimensional surface data reconstruction methods, and can quickly respond to the reconstruction needs of temperature and salinity fields in different regions.

[0172] The above-mentioned three-dimensional temperature and salinity field inversion method based on the ocean temperature and salinity profile, by obtaining one-dimensional temperature and salinity profile data of an observation point, using a pre-trained intelligent reconstruction model to find the nearest triangular net vertex model in the modeling area, calculating the three-dimensional temperature and salinity field data at these vertices, and using an interpolation method to combine these data to reconstruct the accurate three-dimensional temperature and salinity field centered on the observation point, and finally saving the reconstruction result, so as to realize efficient conversion and reconstruction from limited one-dimensional profile data to detailed three-dimensional temperature and salinity field.

[0173] Figure 6 is a schematic block diagram of a three-dimensional temperature and salinity field inversion system 300 based on the ocean temperature and salinity profile provided by the embodiment of the present application. As shown in Figure 6 corresponding to the above-mentioned three-dimensional temperature and salinity field inversion method based on the ocean temperature and salinity profile, the present application also provides a three-dimensional temperature and salinity field inversion system 300 based on the ocean temperature and salinity profile. The three-dimensional temperature and salinity field inversion system 300 based on the ocean temperature and salinity profile includes units for executing the above-mentioned three-dimensional temperature and salinity field inversion method based on the ocean temperature and salinity profile, and the system can be configured in a server. Specifically, please refer to Figure 6 , the three-dimensional temperature and salinity field inversion system 300 based on the ocean temperature and salinity profile includes an acquisition unit 301, a loading unit 302, a searching unit 303, a calculation unit 304, an interpolation unit 305 and a saving unit 306.

[0174] The acquisition unit 301 is used to acquire the observation profile file corresponding to the observation point and extract the key information, wherein the key information includes longitude and latitude, time and temperature and salinity profile, and the temperature and salinity profile includes temperature profile and salinity profile; the loading unit 302 is used to load the pre-trained intelligent reconstruction model; the searching unit 303 is used to search for a plurality of loaded intelligent reconstruction models closest to the observation point in the modeling area range when the observation point is located in the preset modeling area range, so as to obtain the intelligent reconstruction model of the triangular net vertex; the calculation unit 304 is used to calculate the data of the ocean three-dimensional temperature and salinity field according to the intelligent reconstruction model of the triangular net vertex, so as to obtain the three-dimensional body data; the interpolation unit 305 is used to calculate the ocean three-dimensional temperature and salinity field centered on the observation profile by using an interpolation method combined with the three-dimensional body data, so as to obtain the three-dimensional temperature and salinity field data of the target point; and the saving unit 306 is used to save the three-dimensional temperature and salinity field data of the target point.

[0175] In an embodiment, the calculation unit 304 is used to input the temperature and salinity profile and the time into the intelligent reconstruction model of the triangular net vertex to calculate the data of the ocean three-dimensional temperature and salinity field, so as to obtain the three-dimensional body data.

[0176] In an embodiment, the calculation unit 304 includes:

[0177] The space feature extraction subunit is configured to extract space features from the temperature-salinity profile in the vertical direction by using a one-dimensional convolution layer to obtain profile features; the time feature extraction subunit is configured to process input time information by using a full connection layer to obtain time features; the integration subunit is configured to integrate the time features and the profile features by using an addition method to obtain an integration result; and the reconstruction subunit is configured to process the integration result by using a full connection layer and reshape the integration result into a 3D field form, convert the 3D field form by using a maximum and minimum value normalization method, and obtain actual temperature and salinity values to form a three-dimensional volume data.

[0178] In an embodiment, the interpolation unit 305 includes:

[0179] The weight coefficient determination subunit is configured to determine a weight coefficient of the observation point in the triangular net by using a barycentric interpolation method; and the interpolation estimation subunit is configured to estimate a three-dimensional temperature-salinity field at the observation point by using an interpolation method based on the three-dimensional volume data and the weight coefficient to obtain three-dimensional temperature-salinity field data of the target point.

[0180] In an embodiment, the interpolation estimation subunit is configured to perform weighted summation on the weight coefficient of each vertex and the corresponding three-dimensional volume data to obtain the three-dimensional temperature-salinity field data of the target point.

[0181] It should be noted that the specific implementation process of the above-mentioned three-dimensional temperature-salinity field inversion system 300 based on the marine temperature-salinity profile and each unit can be clearly understood by those skilled in the art, and can refer to the corresponding description in the foregoing method embodiments. For the convenience and brevity of description, it will not be described here.

[0182] The above-mentioned three-dimensional temperature-salinity field inversion system 300 based on the marine temperature-salinity profile can be implemented in the form of a computer program, which can run on a computer device as shown in the computer device. Figure 7

[0183] Please refer to Figure 7 , Figure 7 is a schematic block diagram of a computer device provided by an embodiment of the present application. The computer device 500 can be a server, wherein the server can be a stand-alone server or a server cluster composed of multiple servers.

[0184] Referring to Figure 7 , the computer device 500 includes a processor 502, a memory, and a network interface 505 connected through a system bus 501, wherein the memory can include a non-volatile storage medium 503 and an internal memory 504.

[0185] ​The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions which, when executed, can cause the processor 502 to perform a method for three-dimensional temperature and salinity field inversion based on ocean temperature and salinity profile.

[0186] The processor 502 is configured to provide computing and control capabilities to support the operation of the entire computer device 500.

[0187] The non-volatile storage medium 503 provides an environment for the computer program 5032 to run, which, when executed by the processor 502, can cause the processor 502 to perform a method for three-dimensional temperature and salinity field inversion based on ocean temperature and salinity profile.

[0188] The network interface 505 is configured to communicate with other devices via a network. Those skilled in the art can understand that, Figure 7 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device 500 to which the scheme of the present application is applied. The specific computer device 500 can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0189] The processor 502 is configured to run the computer program 5032 stored in the memory to implement the following steps:

[0190] Obtain the observation profile file corresponding to the observation point and extract the key information, wherein the key information includes latitude, longitude, time, and temperature and salinity profile, and the temperature and salinity profile includes temperature profile and salinity profile; load a pre-trained intelligent reconstruction model; when the observation point is located within a preset modeling area range, find a plurality of loaded intelligent reconstruction models closest to the observation point within the modeling area range to obtain an intelligent reconstruction model of a triangular mesh vertex; calculate the data of the ocean three-dimensional temperature and salinity field according to the intelligent reconstruction model of the triangular mesh vertex to obtain three-dimensional body data; calculate the ocean three-dimensional temperature and salinity field centered on the observation profile using an interpolation method combined with the three-dimensional body data to obtain three-dimensional temperature and salinity field data of a target point; and save the three-dimensional temperature and salinity field data of the target point.

[0191] The training process of the pre-trained intelligent reconstruction model includes:

[0192] Construct a triangular mesh of the modeling area and construct an intelligent reconstruction model at the vertex; and train the intelligent reconstruction model using a reanalysis data set.

[0193] In an embodiment, when implementing the step of training the intelligent reconstruction model using a reanalysis data set, the processor 502 specifically implements the following steps:

[0194] The HYCOM reanalysis dataset is selected to obtain a training set; corresponding three-dimensional temperature and salinity field data are extracted from the training set according to the geographic coordinates of the vertices of the triangular mesh of the modeling area and matched with the vertex positions to obtain an extraction result; a data set for model training and verification is constructed for each vertex using the extraction result; the intelligent reconstruction model is trained using the training and verification data sets, an MSE loss function, and an optimization algorithm.

[0195] In an embodiment, when implementing the step of calculating data of a marine three-dimensional temperature and salinity field using the intelligent reconstruction model of the triangular mesh vertices, the processor 502 specifically implements the following steps:

[0196] The temperature and salinity profile and the time are input into the intelligent reconstruction model of the triangular mesh vertices to calculate data of a marine three-dimensional temperature and salinity field to obtain three-dimensional volume data.

[0197] The intelligent reconstruction model includes a deep neural network model with a double-branch structure, which includes a temperature and salinity profile processing branch, a time processing branch, a feature integration layer, and a reconstruction module; the temperature and salinity profile processing branch includes a one-dimensional convolution layer; the time processing branch includes a fully connected layer; and the reconstruction module includes a fully connected layer.

[0198] The time processing branch uses a ReLU activation function, the first fully connected layer expands three input features to eight output features, and the second fully connected layer further expands to 48 output features, matching the output dimension of the temperature and salinity profile processing branch.

[0199] In an embodiment, when implementing the step of calculating data of a marine three-dimensional temperature and salinity field using the intelligent reconstruction model of the triangular mesh vertices, the processor 502 specifically implements the following steps:

[0200] A one-dimensional convolution layer is used to extract spatial features from the temperature and salinity profile in the vertical direction to obtain profile features; a fully connected layer is used to process input time information to obtain time features; the time features and the profile features are integrated by addition to obtain an integration result; the integration result is processed by a fully connected layer and reshaped into a 3D field form, and a maximum and minimum value normalization method is applied for conversion to obtain actual temperature and salinity values, forming three-dimensional volume data.

[0201] In an embodiment, the processor 502, when implementing the step of calculating the ocean three-dimensional temperature and salinity field centered on the observation profile by using the interpolation method in combination with the three-dimensional body data to obtain the three-dimensional temperature and salinity field data of the target point, implements the following steps:

[0202] The weight coefficient of the observation point in the triangular net is determined by using the barycentric interpolation method; and the three-dimensional temperature and salinity field at the observation point is synthetically estimated by using the interpolation method based on the three-dimensional body data and the weight coefficient to obtain the three-dimensional temperature and salinity field data of the target point.

[0203] In an embodiment, the processor 502, when implementing the step of synthetically estimating the three-dimensional temperature and salinity field at the observation point by using the interpolation method based on the three-dimensional body data and the weight coefficient to obtain the three-dimensional temperature and salinity field data of the target point, implements the following steps:

[0204] The three-dimensional temperature and salinity field data of the target point is obtained by weighted summation of the weight coefficient of each vertex and the corresponding three-dimensional body data.

[0205] It should be understood that, in the embodiments of the present application, the processor 502 can be a central processing unit (CPU), and the processor 502 can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0206] It can be understood by those skilled in the art that all or part of the processes in the method of the above-mentioned embodiments can be completed by a computer program instructing related hardware. The computer program includes program instructions, and the computer program can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the above-mentioned embodiments of the method.

[0207] Therefore, the present application further provides a storage medium. The storage medium can be a computer-readable storage medium. The storage medium stores a computer program, wherein the computer program is executed by a processor to make the processor execute the following steps:

[0208] Obtaining an observation profile file corresponding to an observation point, and extracting key information, wherein the key information includes longitude, latitude, time, and a temperature-salinity profile, and the temperature-salinity profile includes a temperature profile and a salinity profile; loading a pre-trained intelligent reconstruction model; when the observation point is located in a preset modeling area range, searching for a plurality of loaded intelligent reconstruction models closest to the observation point in the modeling area range to obtain an intelligent reconstruction model of a triangular mesh vertex; calculating data of a marine three-dimensional temperature-salinity field according to the intelligent reconstruction model of the triangular mesh vertex to obtain three-dimensional body data; calculating a marine three-dimensional temperature-salinity field centered on the observation profile by using an interpolation method combined with the three-dimensional body data to obtain three-dimensional temperature-salinity field data of a target point; and saving the three-dimensional temperature-salinity field data of the target point.

[0209] The training process of the pre-trained intelligent reconstruction model includes:

[0210] Constructing a triangular mesh of a modeling area and constructing an intelligent reconstruction model at a vertex; and training the intelligent reconstruction model using a reanalysis data set.

[0211] In an embodiment, when the processor executes the computer program to implement the step of training the intelligent reconstruction model using a reanalysis data set, the processor specifically implements the following steps:

[0212] Selecting a HYCOM reanalysis data set to obtain a training set; extracting corresponding three-dimensional temperature-salinity field data from the training set according to the vertex geographic coordinates of the triangular mesh of the modeling area and matching the vertex positions to obtain an extraction result; constructing a data set for model training and verification for each vertex by using the extraction result; and training the intelligent reconstruction model by using an MSE loss function and an optimization parameter through a back propagation algorithm combined with the training and verification data set.

[0213] In an embodiment, when the processor executes the computer program to implement the step of calculating data of a marine three-dimensional temperature-salinity field according to the intelligent reconstruction model of the triangular mesh vertex to obtain three-dimensional body data, the processor specifically implements the following steps:

[0214] Inputting the temperature-salinity profile and the time into the intelligent reconstruction model of the triangular mesh vertex to calculate data of a marine three-dimensional temperature-salinity field to obtain three-dimensional body data.

[0215] The intelligent reconstruction model includes a deep neural network model with a double-branch structure, the deep neural network model with the double-branch structure includes a temperature-salinity profile processing branch, a time processing branch, a feature integration layer, and a reconstruction module; the temperature-salinity profile processing branch includes a one-dimensional convolution layer; the time processing branch includes a fully connected layer; and the reconstruction module includes a fully connected layer.

[0216] The time processing branch adopts a ReLU activation function, the first fully connected layer expands three input features to eight output features, and the second fully connected layer further expands to 48 output features, matching the output dimension of the temperature and salinity profile processing branch.

[0217] In an embodiment, when the processor implements the step of calculating the marine three-dimensional temperature and salinity field data by inputting the temperature and salinity profile and the time into the intelligent reconstruction model of the triangular mesh vertexes, the processor implements the following steps:

[0218] The one-dimensional convolution layer is used to extract spatial features from the temperature and salinity profile in the vertical direction to obtain profile features; the fully connected layer is used to process the input time information to obtain time features; the time features and the profile features are integrated by addition to obtain an integration result; the integration result is processed by the fully connected layer and reshaped into a 3D field form, and the maximum and minimum value normalization method is applied for conversion to obtain actual temperature and salinity values, forming three-dimensional volume data.

[0219] In an embodiment, when the processor implements the step of calculating the marine three-dimensional temperature and salinity field data by inputting the temperature and salinity profile and the time into the intelligent reconstruction model of the triangular mesh vertexes, the processor implements the following steps:

[0220] The barycentric interpolation method is used to determine the weight coefficient of the observation point in the triangular mesh; based on the three-dimensional volume data and the weight coefficient, the three-dimensional temperature and salinity field at the observation point is comprehensively estimated by the interpolation method to obtain the target point three-dimensional temperature and salinity field data.

[0221] In an embodiment, when the processor implements the step of calculating the marine three-dimensional temperature and salinity field data by inputting the temperature and salinity profile and the time into the intelligent reconstruction model of the triangular mesh vertexes, the processor implements the following steps:

[0222] The weight coefficient of each vertex and the corresponding three-dimensional volume data are weighted and summed to obtain the target point three-dimensional temperature and salinity field data.

[0223] The storage medium can be a U disk, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk, and various computer readable storage media that can store program codes.

[0224] Those skilled in the art can understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been described in general terms in the above description. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0225] In several embodiments provided by the present application, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiments described above are only illustrative. For example, the division of each unit is only a logical functional division, and actual implementation can have another division. For example, multiple units or components can be combined or integrated into another system, or some features can be omitted or not executed.

[0226] The steps in the method embodiments of the present application can be adjusted, combined and deleted in sequence according to actual needs. The units in the system embodiments of the present application can be combined, divided and deleted according to actual needs. In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit.

[0227] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art, or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a number of instructions for causing a computer device (which can be a personal computer, a terminal or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application.

[0228] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A three-dimensional temperature and salinity field inversion method based on ocean temperature and salinity profiles, characterized in that, include: Obtain the observation profile file corresponding to the observation point and extract key information, including latitude and longitude, time and temperature-salinity profile, the temperature-salinity profile including temperature profile and salinity profile; Load the pre-trained intelligent reconstruction model; When the observation point is located within a preset modeling area, several loaded intelligent reconstruction models that are closest to the observation point are searched within the modeling area to obtain the intelligent reconstruction model of the triangulation vertices. The data of the three-dimensional temperature and salinity field of the ocean are calculated based on the intelligent reconstruction model of the vertices of the triangular network to obtain the three-dimensional volume data; The three-dimensional temperature and salinity field of the ocean centered on the observation profile is calculated by using interpolation methods in combination with the three-dimensional volume data, so as to obtain the three-dimensional temperature and salinity field data of the target point; Save the three-dimensional temperature and salinity field data of the target point; The training process of the pre-trained intelligent reconstruction model includes: Construct a triangulation of the modeling region and build an intelligent reconstruction model at the vertices; The intelligent reconstruction model was trained using the reanalysis dataset; Training the intelligent reconstruction model using the reanalysis dataset includes: Select the HYCOM reanalysis dataset to obtain the training set; Based on the geographic coordinates of the vertices of the triangular mesh in the modeling region, the corresponding three-dimensional temperature and salinity field data are extracted from the training set and matched with the vertex positions to obtain the extraction results; The extracted results are used to construct a dataset for model training and validation for each vertex; The intelligent reconstruction model is trained using the MSE loss function and the parameters are optimized by the backpropagation algorithm, combined with the training and validation datasets. The calculation of the ocean's three-dimensional temperature and salinity field data based on the intelligent reconstruction model of the triangulation vertices to obtain three-dimensional volume data includes: The temperature-salinity profile and the time are input into the intelligent reconstruction model of the triangular mesh vertices to calculate the data of the three-dimensional temperature-salinity field of the ocean, so as to obtain the three-dimensional volume data.

2. The three-dimensional temperature and salinity field inversion method based on ocean temperature and salinity profiles according to claim 1, characterized in that, The intelligent reconstruction model includes a dual-branch deep neural network model, which includes a temperature-salinity profile processing branch, a time processing branch, a feature integration layer, and a reconstruction module. The temperature-salinity profile processing branch includes a one-dimensional convolutional layer. The time processing branch includes a fully connected layer. The reconstruction module includes a fully connected layer.

3. The three-dimensional temperature and salinity field inversion method based on ocean temperature and salinity profiles according to claim 2, characterized in that, The time processing branch uses the ReLU activation function.

4. The three-dimensional temperature and salinity field inversion method based on ocean temperature and salinity profiles according to claim 3, characterized in that, The process of inputting the temperature-salinity profile and the time into the intelligent reconstruction model of the triangulation vertices to calculate the ocean's three-dimensional temperature-salinity field data, in order to obtain three-dimensional volume data, includes: Spatial features are extracted from the temperature-salinity profile in the vertical direction using a one-dimensional convolutional layer to obtain profile features; The input time information is processed through a fully connected layer to obtain time features; The time features and the profile features are integrated by addition to obtain the integrated result; The integrated results are processed through a fully connected layer and reshaped into a 3D field. The maximum-minimum normalization method is then applied to convert the data to obtain the actual temperature and salinity values, thus forming three-dimensional volume data.

5. The three-dimensional temperature and salinity field inversion method based on ocean temperature and salinity profiles according to claim 1, characterized in that, The calculation of the three-dimensional ocean temperature and salinity field centered on the observation profile using interpolation methods combined with the three-dimensional volume data to obtain the three-dimensional temperature and salinity field data of the target point includes: The weight coefficients of the observation points within the triangulation network are determined using the centroid interpolation method. Based on the three-dimensional volume data and the weighting coefficients, the three-dimensional temperature and salinity field at the observation point is comprehensively estimated using an interpolation method to obtain the three-dimensional temperature and salinity field data of the target point.

6. The three-dimensional temperature and salinity field inversion method based on ocean temperature and salinity profiles according to claim 5, characterized in that, The step of comprehensively estimating the three-dimensional temperature and salinity field at the observation point using interpolation methods based on the three-dimensional volume data and the weighting coefficients to obtain the three-dimensional temperature and salinity field data of the target point includes: The target point's three-dimensional temperature and salinity field data are obtained by weighting and summing the weight coefficients of each vertex with the corresponding three-dimensional volume data.

7. A three-dimensional temperature and salinity field inversion system based on ocean temperature and salinity profiles, utilizing the three-dimensional temperature and salinity field inversion method as described in any one of claims 1 to 6, characterized in that, include: The acquisition unit is used to acquire the observation profile file corresponding to the observation point and extract key information, wherein the key information includes latitude and longitude, time and temperature-salinity profile, and the temperature-salinity profile includes temperature profile and salinity profile. The loading unit is used to load pre-trained intelligent reconstruction models. The search unit is used to search for several loaded intelligent reconstruction models that are closest to the observation point within the preset modeling area when the observation point is located within the modeling area, so as to obtain the intelligent reconstruction model of the triangular mesh vertex. The computing unit is used to calculate the data of the three-dimensional temperature and salinity field of the ocean based on the intelligent reconstruction model of the vertices of the triangular mesh, so as to obtain the three-dimensional volume data; An interpolation unit is used to calculate the three-dimensional temperature and salinity field of the ocean centered on the observation profile by combining the three-dimensional volume data with the interpolation method, so as to obtain the three-dimensional temperature and salinity field data of the target point. A storage unit is used to store the three-dimensional temperature and salinity field data of the target point.

Images